Witcher
Services /

AI Security Testing

Understand how your AI application handles hostile inputs, sensitive data and the access it has been given.

Discuss your testing needs

What the work covers.

For security teams, AI owners and product teams preparing an enterprise AI application for production or customer review.

AI security testing examines the application around the model: prompts, retrieval, integrations, identities and controls. A model benchmark alone cannot show whether a user can reach another team’s documents or cause a connected tool to act outside its role. Witcher tests those boundaries in the context of your system and intended use.

  • 01Prompt injection and instruction boundariesTest whether direct inputs or instructions hidden in retrieved content can override the application’s intended behaviour.
  • 02Data exposure and retrieval accessExamine whether responses and retrieval respect user permissions, including attempts to obtain sensitive or out-of-scope information.
  • 03Connected tools and integrationsAssess which actions the application can trigger, what constrains those actions and how failures are surfaced.
  • 04Controls under realistic use and misuseTest the configured application against agreed scenarios, recording what was observed and what remains untested.

Evidence your team
can act on.

A scoped security assessment, reproducible findings, supporting evidence and prioritised remediation recommendations. Retesting can be agreed to check fixes against the original conditions.

Scope, assessment criteria, test access and deliverables are agreed before work begins. See our assurance methodology.

Questions before
you start.

Is this the same as a conventional penetration test?

There can be overlap at application and integration boundaries. This work focuses on AI-specific behaviour, including untrusted instructions, retrieval and model-driven actions. The agreed scope should make clear which conventional infrastructure and application tests are also needed.

What do you need to scope the work?

Start with the intended use, architecture, model and retrieval setup, user roles and connected tools. We agree test access, representative scenarios and boundaries before testing begins.

Part of a wider
assurance engagement.

Explore how this work fits into AI Deployment Assurance, or compare all assurance services.

Testing for your
AI system.

Explore specific testing needs within an assurance engagement.

01

AI Security Testing

Understand how your AI application handles hostile inputs, sensitive data and the access it has been given.

02

LLM Red Teaming

Challenge your LLM application with adversarial scenarios before those failures affect users or customers.

03

AI Agent Security Testing

Test what your agent can read, change and trigger, and whether its permissions match the job it has been given.

Scope your AI testing.

Tell us about the system, the access it has and the decision the evidence needs to support.

Talk to Witcher